WorkMonitor · AI Copilot
Six agents watching what you cannot.
A summary you cannot check is a rumour with formatting. Six agents run on your own numbers, and every claim cites the record it came from. Gated on a live LLM key.
Agents run on your consented data with a live LLM key. Every claim cites its source.
Every finding cites the evidence under it, and a manager decides. A call you can defend when it is challenged.
AI Workforce
What the agents do, and what it costs you
Pick one and the screen it lives on opens beside it, so you judge AI Workforce on what it puts in front of a manager rather than on a list of nouns.
How it works
Putting agents to work on your own data
- 1
Establish the baseline
The agents sit on the same analytics as the productivity picture, so "unusual" means unusual for your team rather than unusual against a threshold somebody else chose.
- 2
Put the agents on duty
Six named agents, each on its own per-organisation schedule, reconciled continuously so one tenant’s workload never delays another’s.
- 3
Ground every claim
Summaries cite the record they came from and are scored for grounding, so an answer can go in front of a client or a board. A prompt-injection guardrail sits between the model and your data.
- 4
Measure what they did
The agent ROI ledger records what each run produced. An AI feature that cannot account for itself is a subscription rather than a capability.
A chat box bolted on vs. agents that account for themselves
workmonitor.vsAI as usually bolted on
What it is grounded in
With WorkMonitor
Your consented data, with a grounding evaluation run against the citations before an answer reaches you.
AI as usually bolted on
A system prompt and hope.
Can you check the claim
With WorkMonitor
Evidence-cited summaries, with the record behind each claim one click away.
AI as usually bolted on
A confident paragraph with nothing behind it.
Prompt injection
With WorkMonitor
A guardrail between the model and the data, treated as a threat rather than an edge case.
AI as usually bolted on
Unconsidered, until content in your own data starts issuing instructions.
Autonomy
With WorkMonitor
They surface and draft; your managers decide. Nothing irreversible lands on a person because a model said so.
AI as usually bolted on
Marketed as agents that act, which nobody actually enables.
Proving it was worth it
With WorkMonitor
An ROI ledger recording what each agent run produced, priced against the reviewed work it stood in for.
AI as usually bolted on
A line on the invoice and a feeling.
Scheduling at scale
With WorkMonitor
Per-organisation schedulers on each org’s own cron, reconciled every couple of minutes.
AI as usually bolted on
One global cron, so a large tenant delays a small one.
Every capability in AI Workforce, and exactly where each one stands
- Six named agents
- AI Copilot (evidence-cited summaries + ask)
Ask AI
Ask AI about your AI workforce
The ROI ledger and the grounding evals, in plain language. This is the AI reporting on itself, citations included.
Pick a question above and WorkMonitor AI will answer from your team's real numbers.
By team
Who the agents report to
Use cases
What the agents get put on
Each one opens on the page written for that job — the same capabilities, read from the work rather than from the product.
- Catch anomalies with smart alertsAnomaly detection against your own baseline, six named agents on specific jobs, evidence-cited summaries, and delivery to a channel your team already reads.
- Score productivity & focusOne score per person, team or company, with the breakdown behind every one. Your categories rather than our defaults, and a capacity signal that names who is running hot.
- Manage remote & hybrid teamsOne live board across every zone, attendance in each person’s local day, capacity bands that name who is drowning, and a rollout that never touches a machine.
Straight answers
The questions we would ask in your position
Every answer here is the one you would get on a call. Open as many as you like; they stay open, so two can be held side by side.
No, by design, and that is a control rather than a limitation. The six agents draft, detect and flag; a manager makes the call, so nothing consequential lands on a person on a model’s say-so. There is no configuration in which an agent acts silently on your workforce data.
Every summary or answer the AI Copilot produces carries citations back to the records it drew from, so you check a claim before you act on it. No citation, no claim. Grounding evals then score outputs against that source data continuously, so drift shows up as a falling score rather than as a decision you have to unpick six weeks later.
No. It answers from a bounded snapshot of the same per-member rollup figures the analytics screens already show, and it hands that snapshot back with the answer. Every fact in it is numbered to the figures it was drawn from, and a sentence it cannot tie back to the snapshot does not get written. The six named agents run on the same data under the same rule.
A dedicated prompt-injection guardrail screens the content agents read before it can steer them, and it ships alongside the grounding evals in the trust layer. The advisory-only ceiling is the backstop: even a compromised suggestion is still a suggestion, and a human still reviews it.
The agent ROI ledger prices every action and sets it against the reviewed work it produced, so the return is a number you can audit rather than a feeling. When an agent stops earning its keep, the ledger is where you find out first — and switching it off is one decision, not a renegotiation.
Take these with you
The software is the easy part of a rollout
Here is what we would send a manager doing one for the first time: how to read a productivity number, what to say to a remote team before anything is installed, and a policy you can adopt as written.
Point it at one team for a week.
Create the account, put the agent on a handful of desks, and leave it alone. On Friday you read the week instead of reconstructing it: hours against their projects, focus and idle per person, and the timesheets already filled in.
Free for two seats. No card, and no sales call to sit through.